In the last two decades, hyperspectral and multispectral imaging have become more popular and useful. These terms are often used interchangeably, but they represent two different imaging methods with distinct application areas. Hyperspectral and multispectral imaging offer advantages over traditional machine vision imaging methods that use light in the visible spectrum (400 to 700nm). These benefits are accompanied by an increase in system complexity, including lighting, filters, and optical design.
In most machine vision applications the illumination captured and used by the sensor falls within the visible spectrum. The human eye is only able to detect light in this part of spectrum, which ranges from 400nm violet (or dark red) up to 700nm deep red. Sensors and imaging lens assemblies have a peak spectral sensitivity around 550nm. Quantum efficiency is the camera sensor’s ability to convert photons in an electric signal. This efficiency drops significantly as the wavelengths of ultraviolet and near-infrared increase. Hyperspectral Imaging (HSI) can be defined as a way to capture images that include information from a wider portion of electromagnetic spectrum. This can begin with UV light and extend to the visible spectrum before ending in near-wave or shortwave infrared. The extended wavelength range can reveal material properties that would not be apparent otherwise.
The output of machine vision sensors is a grayscale array of values that results in a two-dimensional image of an object within the viewing area. This is primarily used for feature recognition to sort, measure, or locate objects. Unless optical filters are used, the vision system will not know what wavelengths are being used to illuminate an object. The Bayer Pattern RGB filter is the exception, however, even in this case, each pixel can only accept light from a limited range of wavelengths. In a hyperspectral picture, each pixel is associated with coordinates, wavelengths and signal intensities. HSI, or imaging spectroscopy, is sometimes called HSI.
A spectrometer is a device that collects light from one source. A spectrometer is used to detect substances which scatter and reflect certain wavelengths or the composition of materials based on fluorescent emissions. The HSI system extends this technology by assigning locational data to light spectrums. Hyperspectral and multispectral imaging do not produce a 2D picture, but rather a hyperspectral cube.
Each mode has its own advantages and disadvantages . The whiskbroom is a scanning method that uses a single spatial coordinate to acquire spectral data. This method offers the highest spectral resolution but also requires that the system scans the target area along both the x- and y-axes. Pushbroom is a line scan data capture method in which only one axis is needed to move the pixels over the area. These pushbroom systems are compact, light, and have a higher signal-to-noise ratio. 1 It is important to get the timing of exposures right when using this HSI technique. A wrong exposure time will result in inconsistent saturation of the spectral bands or underexposure. Plane scanning involves capturing multiple images at each wavelength interval to produce the spectral depth in the hyperspectral cube. This method of capture does not require the translation of the entire system or sensor, but it is important that the subject not move during the acquisition. Otherwise the accuracy of the positional and spectrum information will be compromised. Single shot, or snapshot, is the fourth and most recent mode of hyperspectral and multispectral imaging acquisition. Single shot imagers collect the entire hyperspectral cube in a single integration period.
The MSI system is very similar to a hyperspectral one, but there are some key differences. An MSI focuses on a few pre-selected wavebands based upon the application, as opposed to an HSI which collects data in a continuous manner. RGB sensors are a good example of this concept, even though they’re not directly comparable or an exact comparison. The Bayer pattern is applied to RGB sensors, which consists of red, blue, and green filters. These filters allow specific wavelengths to be absorbed into pixels, while attenuating the rest. Bandpass filters are characterized by transmission bands between 400-700nm with a slight overlap in spectral range. Images are rendered in false colors to simulate what the eye sees. In multispectral applications, wavelength bands are narrower and larger. Wavebands can be tens or hundreds of nanometers wide and not only part of the visible spectrum. According to the application, thermal wavelengths, such as mid-wave infrared, UV and NIR can also have their own channels. 4
MSI is viewed by some as a less effective version of HSI with a lower spectral resolving power. Both technologies have their advantages, which make them suitable for different applications. HSI works best for applications that are sensitive to small differences in signals along a continuum spectrum. A system that samples large wavebands may miss these small signals. Some systems, however, require that significant portions of electromagnetic spectrum be blocked in order to selectively capture the light . Other wavelengths may cause significant noise, which could ruin measurements and observations. If there is less information in the datacube, image capture, analysis, and processing can be done faster.
The number of applications that use HSI or MSI is growing. Since decades, remote sensing – aerial imaging of earth’s surface using unmanned aerial vehicles and satellites – has relied heavily on HSI and MSI. The spectral photography allows for a clear view of the earth’s surface, even through different clouds and Earth’s atmosphere. This technology is used to study archeological sites, monitor population changes, and observe geological changes. HSI and MSI technology are also becoming increasingly important in studying the environment. The data collected can include deforestation and ecosystem degradation, as well as carbon recycling and more erratic weather patterns. Researchers can use this information to build predictive models for the global eco-system, which is used to drive many environmental initiatives to combat climate change and the human impact on nature.
In the medical field, it is no different. hyperspectral and multispectral imaging allows doctors to perform non-invasive skin scans that detect malignant or diseased cells. Certain wavelengths penetrate deeper into the skin and provide a better understanding of the patient’s condition. Under the right stimulation, cancerous cells and other diseases will fluoresce or absorb light. Doctors no longer have to guess based on the symptoms described by patients and what they see. Advanced systems can automatically record and interpret spectral data. This allows for a much faster diagnosis and treatment.
These technologies are used in many different areas. Life sciences, pharmaceuticals and healthcare are some of the more specific ones. These tools are particularly helpful for farmers, as they can determine the growth rate of their crops. Tractors and drones equipped with spectral imaging sensors can scan fields at lower altitudes. Farmers then analyze the spectral properties of the captured images. These characteristics can be used to determine whether the plants are healthy, the condition of the soil or the presence of an infection. The information is accompanied by unique spectral marks that can be analyzed and used to optimize the production of the produce.
Despite the fact that HSI/MSI has a wide range of applications, and is growing rapidly, industry adoption of this technology is slow due to limitations. These systems are currently significantly more expensive than other machine vision components. Sensors must be more sophisticated, have greater spectral sensitivity and be calibrated precisely. It is common for sensor chips to require substrates other that silicon, as it only has a sensitivity range of 200-1000nm. Light up to 2600nm can be collected using Indium arsenide, gallium arsenide or indium galium arsenide. In order to capture images from the NIR up to the MWIR range, you will need a microbolometer or other sensor with a longer wavelength. These systems require larger sensors and pixels than most machine vision sensors in order to achieve the required spatial resolution and sensitivity.
The challenge comes when matching these high-end sensor with the right optical components. The lenses for these cameras need to be designed in a way that is compatible with the vast range of wavelengths and temperature variations. This design requires more optical components, increasing the cost and weight of the system. For broadband color correction, elements will have to have different dispersive and refractive properties. Different glass types have different thermal and mechanical properties. It is important to use anti-reflection multi-layer coatings on each lens after selecting the glasses with the correct internal transmission spectrum. This will ensure maximum light transmission. In these conditions, the lens design for hyperspectral and multispectral imaging is a tedious process that requires great skills. In certain application areas, it is also necessary that the lens assemblies be athermal. This will ensure that the system functions the same on the ground as well as in the upper atmosphere.
The future development goal is to make HSI/MSI systems smaller, more affordable and easier to use. These improvements will encourage new markets to use the technology and help advance those markets that are already using it.
To learn more, please feel free to contact us at info@superiorcctv.com.
English